$\newcommand{\set}[1]{\{#1\}}$ $\newcommand{\mc}{\mathcal}$

Definitions and Context

Let $(X, \mc X, \mu, T)$ be an invertible measure preserving system and let $A$ be a measurable subset of $X$ with $\mu(A)>0$. By Poincare recurrence we know that for almost all points $x\in A$ there is a positive integer $k$ such that $T^kx\in A$. Thus we may define a map $r_A:A\to \mathbb N$ by writing $$ r_A(x) = \min\set{k\geq 1:\ T^kx\in A} $$ So $r_A$ is well-defined almost everywhere in $A$. Also define a map $T_A:A\to A$ by writing $$ T_Ax= T^{r_A(x)}x $$ Let $\mu_A$ be a measure on $A$ obtained by restricting $\mu$ to $A$ and normalizing, that is, $\mu_A(B)=\mu(B)/\mu(A)$ for any measurable subset $B$ of $A$.

The Problem

Lemma 2.43 in Einsiedler and Ward's Ergodic Theory with a view towards Number Theory states that $T_A$ preserves the measure $\mu_A$. What the proof given in the book shows is that for a measurable subset $B$ of $A$ we have $\mu_A(T_A(B))=\mu_A(B)$. I do not see how this shows that $T_A$ is measure preserving. What one needs to show is that for any measurable subset $B$ of $A$ we have $\mu_A(T_A^{-1}(B))=\mu_A(B)$, which is equivalent to showing that $\mu(T_A^{-1}(B))=\mu(B)$.

Here is what I tried. Define, as in the book, $A_n=r_A^{-1}(n)$. Thus the $A_i$'s partition $A$. Let $B$ be an arbitrary measurable subset of $A$. Then we have $$ T_A^{-1}(B) = \bigsqcup_{n\geq 1} \set{x\in A_n:\ T_Ax\in B} = \bigsqcup_{n\geq 1} \set{x\in A_n:\ T^nx\in B} $$ Thus $$ T_A^{-1}(B) = \bigsqcup_{n\geq 1} A_n\cap T^{-n}(B) $$ So we have $$ \mu(T_A^{-1}(B)) = \sum_{n\geq 1} \mu(A_n\cap T^{-n}(B)) $$ I am stuck here.


This fact is general for any invertible measure preserving system:

Let $(X,\mathcal{B},\mu,T)$ be such that $T$ is invertible and measurable with $T^{-1}$ also measurable. Then $\mu(T^{-1}(A))=A$ for all $A\in\mathcal{B}$ if and only if $\mu(TA) = \mu(A)$ for all $A\in\mathcal{B}$

In fact we only need one direction of this theorem, that $\mu(TA)=\mu(A)$ for all $A\in\mathcal{B}$ implies that $\mu(T^{-1}(A))=A$ for all $A\in\mathcal{B}$. This direction alone holds even if $T$ is not invertible, but is surjective. We will prove that in a minute, but this is what we need for your question.

Back to your question: Let $B$ be a measurable subset of $A$. Then since $T$ and $r_A$ are measurable, we have that $C:=T_A^{-1}(B)$ is also a measurable subset of $A$.

The proof given in the book shows that $\mu_A(T_A C) = \mu_A(C)$. Since $T_A$ is surjective (see proof below) we have $$T_A(C) = T_A T_A^{-1}(B)=B$$ which implies that $\mu_A(B)=\mu_A(T_A^{-1}(B))$ as required.

The proof that $T_A$ is surjective goes as follows: Consider the measure-preserving system $(X,\mathcal{B},\mu,T^{-1})$ and the same measurable set $A$. By poincare recurrence theorem for almost all $a\in A$ there exists $n$ such that $T^{-n}(a)\in A$. Let $b:=T^{-n}(a)$ for the minimal $n$, then $T_A(b)=a$. This shows that $T_A$ is essentially surjective which is enough for our purposes.

| cite | improve this answer | |
  • 1
    $\begingroup$ That's a very nice way to do it. Thanks. I have just one question. The general result that you mention about $\mu(TA)=\mu(A)$ implying $T$ is measure preserving given that $T$ is surjective, isn't one assuming that the image of any measurable set under $T$ is also measurable? (I see that in our case this is not a problem.) $\endgroup$ – caffeinemachine Apr 21 '19 at 11:47
  • $\begingroup$ @caffeinemachine Yes, I guess you have to assume that otherwise $\mu(TA)$ makes no sense. $\endgroup$ – Yanko Apr 21 '19 at 11:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.